Systematic meta-review: diagnostic accuracy of colon capsule endoscopy for colonic neoplasia with umbrella meta-analysis

Abstract

Background:

Colorectal cancer (CRC) incidence is rising globally, intensifying pressure on endoscopy services. Colon capsule endoscopy (CCE) offers a non-invasive alternative. Despite several systematic reviews showing reasonable polyp detection rates, clinical scepticism remains.

Objectives:

This meta-review and umbrella meta-analysis aim to synthesise evidence on CCE’s diagnostic accuracy in polyp and CRC detection, using CT colonography or colonoscopy as the reference standard.

Methods:

We conducted a systematic search of EMBASE, MEDLINE and PubMed for systematic reviews evaluating the diagnostic accuracy of CCE in detecting polyps and CRC. A qualitative thematic review and synthesis were conducted following PRISMA guidelines. A bivariate generalised linear mixed model with random effects was used for pooled diagnostic accuracy estimates, and meta-regression was performed using restricted maximum likelihood estimation.

Results:

Nine systematic reviews encompassing 28 unique studies (3472 participants) were included. For polyps of any size, the pooled per-patient sensitivity was 0.79 (95% CI: 0.69–0.86), specificity was 0.77 (95% CI: 0.71–0.82), and the area under the curve (AUC) was 0.81. For polyps ⩾6 mm, sensitivity and specificity were 0.80 and 0.87 (AUC 0.81), and for polyps ⩾10 mm, 0.88 and 0.95 (AUC 0.95), respectively. Second-generation CCE (CCE2) improved diagnostic accuracy across all polyp sizes. For polyps of any size, CCE2 achieved a sensitivity of 0.90, specificity of 0.81 and AUC of 0.82. For polyps ⩾ 6 mm and ⩾10 mm, AUCs were 0.92 and 0.94, respectively. CCE2 showed high sensitivity for detecting any polyp size and polyps ⩾6 mm, with low heterogeneity (p > 0.05, I² < 25%). CRC detection sensitivity was 0.96 (95% CI: 0.73–1.00) after excluding cases where the capsule failed to reach the rectum due to battery exhaustion.

Conclusion:

CCE2 has high diagnostic accuracy for polyps and colorectal cancer detection. While technical challenges persist, CCE2 shows promise as a complementary diagnostic tool to help address the increasing demands for endoscopy services.

Plain language summary

Colon capsule endoscopy (CCE) is a camera pill that patients swallow to check the inside of their colon without needing a traditional colonoscopy. Although this method is less invasive, it hasn’t been widely adopted, partly because doctors are unsure how accurate it is in spotting problems like polyps (small growths) or bowel cancer. This study reviewed and combined data from nine major research reviews involving 3,472 patients. The goal was to find out how good CCE is at detecting polyps and colorectal cancer (bowel cancer) compared to standard tests like colonoscopy and CT colonography. The results showed that second-generation CCE devices (CCE2) performed well. For larger polyps (over 10 mm), the camera pill correctly identified them in 88% of cases and correctly ruled them out in 95% of cases. When it came to detecting bowel cancer, the accuracy was even higher, around 96%, as long as the capsule finished the test (seen the whole colon) before the battery ran out. The newer version of the capsule (CCE2) was better than older models, especially in detecting both small and large polyps. This suggests that, with current technology, CCE can be a reliable tool for detecting bowel problems early. In summary, CCE2 offers a promising alternative for bowel screening, particularly when regular colonoscopy is delayed or difficult. It could help ease pressure on healthcare services while still delivering accurate results.

Introduction

Since its introduction in 2006, colon capsule endoscopy (CCE) has increasingly contributed to the diagnostic workload for colonic diseases, serving as a second-line investigative modality in endoscopy services. Colorectal pathology, particularly colorectal cancer (CRC) and polyps, remains a primary global health concern, emphasising the need for effective screening strategies to improve patient outcomes. ¹ As a non-invasive alternative to conventional colonoscopy, CCE enables complete colonic visualisation without sedation or hospital attendance, especially with the recent advancement in telemedicine, enhancing patient comfort and accessibility. ² Compared to its predecessor (CCE-1), the second-generation colon capsule endoscopy system (CCE-2) offers significantly improved diagnostic performance through several key technological upgrades. These include enhanced image resolution, an expanded field of view from 156° to 172° per camera (providing a combined 344° coverage), and the implementation of adaptive frame rates ranging from 4 to 35 frames per second (fps), adjusting dynamically to capsule transit speed as an advancement over the fixed 4–14 fps range in CCE-1. In addition, CCE-2 benefits from an extended battery life of up to 12 h, compared to approximately 10 h in CCE-1. ³ Moreover, integrating artificial intelligence (AI) into CCE is gaining momentum, further expanding its diagnostic potential and clinical appeal.^4,5

Despite these advancements, the adoption of CCE remains challenged by factors affecting diagnostic performance. These include needing a more stringent bowel preparation regimen to achieve adequate cleansing quality, the lengthy procedure time and the risk of capsule retention before battery exhaustion. Standardised protocols and ongoing technological refinements are crucial to optimising CCE’s clinical utility and global uptake. Recent systematic reviews and meta-analyses provide compelling evidence supporting the diagnostic accuracy of CCE in polyp detection. However, scepticism persists, with concerns over its reliability slowing its adoption. Uncertainty surrounding diagnostic accuracy remains one of the primary barriers to widespread implementation. This study aims to conduct a systematic review and umbrella meta-analysis of existing systematic reviews and meta-analyses to evaluate the diagnostic accuracy of CCE for polyp and CRC detection. In addition, it seeks to identify factors influencing diagnostic performance, consolidating findings to inform clinical applications and future research, particularly in the context of emerging AI technologies.

Methods

The study protocol was designed based on the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guideline. ⁶ The primary objective was to systematically synthesise evidence from systematic reviews and meta-analyses evaluating the diagnostic accuracy of CCE in polyp detection, utilising both qualitative thematic analysis and quantitative umbrella meta-analysis. The secondary objective was to assess the impact of key covariates, including CCE models, completion rate, bowel preparation regimen, bowel cleansing adequacy and CCE indications on polyp diagnostic accuracy.

Eligibility criteria

The search included all systematic reviews evaluating the diagnostic accuracy of CCE in polyp detection, with no language restrictions. Eligible reviews were required to provide a clear comparison between CCE and either CT colonography or colonoscopy. There were no restrictions on the number of primary studies, CCE indications, study types (including conference abstracts), or CCE capsule models. In addition, polyp diagnostic accuracy did not need to be the primary outcome in the unique primary studies included in these systematic reviews.

Information sources

A systematic search was conducted using PUBMED, EMBASE and MEDLINE from 1 January 2006 to 31 December 2024. Additional publications were hand-searched using the references of the extracted studies. The search employed a combination of Medical Subject Headings (MeSH) and free-text terms such as ‘polyp’, ‘adenomas’, and ‘Colorectal cancer’ (see Supplemental Appendix A, Table A1). The search strings used for each database are available in Supplemental Appendix A.

Study selection

The titles and abstracts of all retrieved studies were screened by three authors (I.L., P.C. and A.K.). The inclusion criteria were:

• (a) Study types: Systematic reviews with or without meta-analyses evaluating the diagnostic performance of CCE in polyp and CRC detection.

• (b) Primary outcomes: diagnostic accuracy encompassed sensitivity, specificity, diagnostic odds ratios (DORs) and area under the curve (AUC) for detecting polyps and CRC.

• (c) Populations: Studies involving average-risk screening and symptomatic populations with indications for lower gastrointestinal examination.

The Exclusion criteria were the lack of extractable diagnostic metrics or raw data for diagnostic metrics calculation and the use of a small bowel capsule.

Data compilation

The final selected systematic reviews were evaluated, and a citation matrix was constructed (see Supplemental Appendix A, Table A1). This matrix was used to calculate the Corrected Covered Area (CCA), which served as a key metric in assessing study overlap within the umbrella review. ⁷ The CCA interpretation thresholds are classified as follows: values below 5% indicate slight overlap with minimal risk of bias; 6%–10% represents moderate overlap; 11%–15% reflects high overlap; and values exceeding 15% suggest very high overlap, posing a serious risk of duplication bias in meta-reviews.^8,9 When high overlap (CCA > 10%) was detected, all unique primary studies were identified across the included systematic reviews to mitigate redundancy. Additional data extraction and meta-analysis were subsequently conducted by I.L. using the non-duplicated primary studies. Extracted variables included per-patient diagnostic metrics, heterogeneity, polyp size, capsule type, bowel preparation and booster regimens, bowel cleansing quality and capsule excretion rate before battery exhaustion (see Supplemental Material Appendix Tables A2 and A3). ⁹

Risk of bias

The methodological quality of included systematic reviews was assessed using the AMSTAR-2 (A Measurement Tool to Assess Systematic Reviews) framework. ¹⁰ The quality assessment of the systematic reviews was classified as low, low-moderate, moderate, moderate-high and high quality, as per the authors’ instructions. ¹⁰

Statistical analysis

Two-by-two contingency tables, reporting per-patient true positives (TP), true negatives (TN), false positives (FP) and false negatives (FN), were extracted from the summary tables of meta-analyses and, when unavailable, from individual primary studies. Key diagnostic accuracy metrics were derived from these matrices, including sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV). ¹¹ A diagnostic test accuracy meta-analysis was conducted using a bivariate logistic regression model with random effects (bivariate GLMM).^12,13 This model, based on TP, TN, FP and FN data pairs, was used to estimate pooled sensitivity, specificity, DOR, positive likelihood ratio (PLR), negative likelihood ratio (NLR) and the AUC through a summary receiver operating characteristic (sROC) curve. The sROC curves provided a comprehensive evaluation of diagnostic performance. The PLR indicates the probability of CCE correctly identifying a patient with polyps, whereas the NLR reflects the likelihood that a negative CCE result still corresponds to a true case. The AUC served as a global measure of diagnostic accuracy, categorised as low (AUC < 0.7), moderate (0.7 ⩽ AUC < 0.9) or high (AUC ⩾ 0.9). ¹⁴ Variance in logit-transformed sensitivity and specificity was evaluated to assess heterogeneity and forest plots were used across studies to visualise and explore potential sources of heterogeneity. The umbrella meta-analysis was performed using the ‘lme4’ ¹³ ‘metafor’ ¹⁵ , ‘mada’ (version 0.5.11) ¹⁶ , ‘lmtest’ ¹⁷ R packages within R software (R Core Team, 2025). ¹⁸

Meta-regression and subgroup analysis

Meta-regression analyses were performed to assess the impact of study-level covariates on pooled diagnostic accuracy estimates, including bowel cleansing adequacy, capsule endoscopy generation (CCE-1 vs CCE-2), capsule excretion rate and polyp size thresholds. Variance estimates were computed for sensitivity and specificity to account for within-study variability. Restricted maximum likelihood (REML) estimation within the ‘metafor’ package ¹⁵ was used to fit random-effects meta-regression models, ensuring variance estimation across heterogeneous study designs.

A subsequent subgroup analyses were conducted by stratifying studies based on key diagnostic accuracy determinants, including study type, polyp size categories and capsule generation. Pooled estimates for sensitivity, specificity, PLR, NLR, DOR and AUC were computed separately for each subgroup, allowing for a more refined assessment of potential sources of heterogeneity. Model predictors included completion rate, adequate bowel preparation, study design, polyp size threshold and capsule generation, incorporated as moderators in the meta-regression models.

Heterogeneity was quantified using between-study variance estimates (τ²) and the proportion of variance due to heterogeneity (I²). Sensitivity analyses were conducted by excluding outliers and recalculating pooled estimates. Meta-regression models were visualised using predicted logit-sensitivity and logit-specificity plots, illustrating the relationship between study-level covariates and diagnostic performance. The results of these models provided insights into the factors influencing diagnostic accuracy variations and facilitated a comprehensive interpretation of the pooled estimates.

An effective sample size (ESS) funnel plot was constructed alongside Deeks’ funnel-plot asymmetry test to assess potential systematic differences among the included studies and evaluate publication bias. This regression-based method examines the presence of publication bias by plotting the DOR against the inverse square root of the ESS.

Results

Literature search and study selection

A total of 424 references were identified through the literature search across three databases (see Figure 1). One study was excluded as it focused solely on IBD rather than polypoidal lesions. In addition, one study was identified through manual citation screening. Ultimately, nine systematic reviews met the inclusion criteria, of which eight included a meta-analysis and were incorporated into the umbrella meta-analyses.^3,19–26

Figure 1.

PRISMA flow chart.

The study overlap assessment using the CCA citation matrix is presented in Table A2 in the Supplemental Appendix. A total of 82 studies were included across nine systematic reviews, but only 31 were unique primary studies. The 100% CCA indicates a high overlap among the systematic reviews. Given this, further data extraction was conducted from the 31 unique primary studies included in the nine systematic reviews, with analysis performed by I.L. (see Supplemental Appendix Table A3). Subsequently, three conference abstracts were removed to avoid duplication, as their corresponding subsequent full-text studies were already included.^27–29 In addition, three primary studies were excluded from the meta-analysis due to a non-diagnostic accuracy focus, ³⁰ inclusion of only CRC patients, ³¹ and insufficient extractable data. ³²

Study characteristics

Table 1 summarises the pooled diagnostic metrics from each systematic review and meta-analysis included in this study. Only one review was found to have a moderate risk of bias ¹⁹ due to a partial literature search and the absence of quality rating and publication bias assessment. The remaining studies had a low risk of bias (see Supplemental Table A4). The most common limitations were incomplete literature searches, where search terms were not as comprehensive, and the lack of publication bias assessment, often due to the limited number of included studies available at the time of publication. Although Deek’s regression test did not indicate statistically significant publication bias (t = 0.77, df = 12, p = 0.4564), the estimated bias was 7.23 (SE = 9.39), suggesting high uncertainty in the bias estimate. In addition, no apparent asymmetry was observed in the funnel plot (see Figure A2 in the Supplemental Appendix), indicating no firm evidence of publication bias.

Table 1.

Polyp detection result from different systematic reviews and meta-analysis.

Author and location	Number of studies and participants	Type of included studies	Overall pooled sensitivity (any polyp size)	Sensitivity Heterogeneity (any size)	Overall pooled Specificity (any polyp size)	Specificity Heterogeneity (any polyp size)	Pooled AUC (any size)	Pooled Sensitivity for polyp ⩾ 10 mm	Pooled Specificity for polyp ⩾ 10 mm	Pooled AUC Polyp ⩾ 10 mm	Pooled Sensitivity for polyp ⩾6	Pooled Specificity for polyp ⩾6	Pooled AUC Polyp ⩾ 6 mm	CCE type
Rokkas et al. 2010 (Greece) 19	7 studies (n = 626)	Full article and abstract	0.73 (0.68–0.77)	Q = 0.81, df = 3, p = 0.85, I2 = 0%	0.89 (0.81–0.94)	Q = 18.6, df = 3, p = .0003, I2 = 83.9%	0.797	–	–	–	0.69 (0.62–0.75)	0.86 (0.82–0.90)	–	CCE1
Spada et al. 2010 (Italy) 20	8 studies (n = 837)	Full article and abstract	0.71 (0.66–0.76)	Q = 8.04, df = 7, p = 0.33 I2 = 12.89	0.65 (0.66–0.83)	Q = 15.01, df = 7, p = 0.04 I2 = 53.37	–	–	–	–	0.68 (0.56–0.79)	0.82 (0.77–0.85)	–	CCE1
Spada et al. 2016 (Italy) 3	14 studies (n = 2420)	Full article and abstract	–	–	–	–	–	0.81 (0.66–0.90)	0.96 (0.94–0.98)	0.94	0.74 (0.61–0.84)	0.87 (0.81–0.91)	–	CCE1 and CCE2
Vuik et al. 2020 (Netherlands) 21	13 studies (n = 2485)	Full article and abstract	No meta-analysis	No meta-analysis	No meta-analysis	No meta-analysis	No meta-analysis	No meta-analysis	No meta-analysis	No meta-analysis	No meta-analysis	No meta-analysis	No meta-analysis	CCE1 and CCE2
Alihosseini et al. 2020 (Iran) 22	8 studies (n = 1238)	Full article	0.93 (0.84–0.97)	Q = 0.37, df = 1, p = 0.546 I2 = 0.0	0.66 (0.48–0.81)	Q = 0.00, df = 1, p = 0.9442 I2 = 0.0	–	0.84 (0.76–0.89)	0.96 (0.94–0.97)	0.96	0.84 (0.80–0.88)	0.88 (0.85–0.90)	0.91	CCE2
Kjølhede et al. 2020 (Denmark) 23	12 studies (n = 2199)	Full article	0.85 (0.73–0.92)	p = 0.98	0.85 (0.70–0.93)	p = 0.98	–	0.87 (0.82–0.90)	0.95 (0.92–0.97)	NA	0.87 (0.83–0.90)	0.88 (0.75–0.95)	–	CCE2
Ali 2021 (USA) 24	5 studies (n = 1305)	Full article	–	–	–	–	–	0.86 (0.8–0.91)	0.96 (0.92–0.98)	NA	0.86 (0.82–0.91)	0.88 (0.72–0.96)	–	CCE2
Möller et al. 2021 (Germany) 25	9 studies (n = 2328)	Full article	–	–	–	–	–	0.87 (0.83–0.90)	0.95 (0.91–0.97)	NA	0.87 (0.83–0.90)	0.87 (0.76–0.93)	–	CCE2
Sulbaran et al. 2021 (Brazil) 26	8 studies (n = 1602)	Full article and abstract	–	–	–	–	–	0.88 (0.82–0.93)	0.95 (0.94–0.97)	0.97	0.88 (0.84–0.91)	0.94 (0.92–0.95)	0.96	CCE2

Quantitative analysis

The 28 unique prospective primary studies, published between 2006 and 2024, reported 3,472 enrolled patients. A summary of these studies is provided in Table A2 (Supplemental Appendix). For 10 primary studies,^33–43 raw data extraction was required for TP, TN, FP and FN. To ensure consistency in analysis, polyp size categories were standardised into ‘Any polyp size’, ‘Polyp ⩾ 6 mm’ and ‘Polyp ⩾ 10 mm’, the commonly recognised thresholds for post-polypectomy surveillance in colonoscopy.^44,45 These thresholds were also previously adopted in the included systematic reviews (see Table 1).

Bowel cleansing, capsule excretion rate before battery exhaustion and capsule retention

Both bowel cleansing and capsule excretion before battery exhaustion are critical for diagnostic accuracy and limiting colonoscopy conversion rates. The proportion of adequate bowel cleansing varied widely among included studies, ranging from 40% to 90%, while capsule excretion before battery depletion ranged from 64% to 100%. There was considerable variation in bowel preparation protocols and booster regimens across studies. However, polyethylene glycol (PEG) was universally used, with differing volumes. Only one study abstract failed to report details on bowel preparation or excretion rates. As for booster agents, sodium phosphate-based laxatives were used in 77.5% (24/31) of studies, alone or in combination with PEG, magnesium citrate, or gastrografin. Regarding capsule retention, Spada et al. was the only systematic review that explicitly reported the absence of any retention events, ²⁰ while the remaining reviews did not specify retention outcomes. Among the primary studies, Ota et al., ³¹ Morgan et al. ³⁶ and Akyüz et al. ³⁵ clearly documented that no capsule retention occurred. Several other studies reported no adverse events, although they did not directly address retention. Collectively, these findings support the conclusion that capsule retention is a rare complication in CCE.

Diagnostic accuracy in polyp detection

Table 2 presents the diagnostic accuracy of polyp detection across various subgroups, including polyp size thresholds and capsule models. Given the significant heterogeneity across studies, GLMM and univariate random-effects meta-analysis were performed to estimate pooled diagnostic metrics. The results of both analyses and heterogeneity estimates are summarised in Table 3. Figures 2 and 3 illustrate forest plots of the pooled sensitivity and specificity stratified by polyp size subgroups.

Table 2.

Overall and subgroup analysis for diagnostic accuracy of CCE in polyp detection using GLMM.

Overall and subgroup analysis	Pooled sensitivity (95% CI)	Pooled specificity (95% CI)	Pooled PLR (95% CI)	Pooled NLR (95% CI)	Pooled DOR (95% CI)	SROC-AUC (95% CI)
Any Polyp size (n = 14)	0.79 (0.69–0.86)	0.77 (0.71–0.82)	3.36 (2.54–4.19)	0.28 (0.17–0.38)	12.21 (5.39–19.03)	0.81 (0.47–0.96)
Polyp size ⩾6 mm (n = 18)	0.80(0.72–0.86)	0.87 (0.82–0.91)	6.07 (4.03–8.11)	0.22 (0.15–0.30)	27.04 (11.94–42.13)	0.81 (0.48–0.95)
Polyp size ⩾10 mm (n = 14)	0.88 (0.79–0.94)	0.95 (0.90–0.97)	16.80 (5.38–28.21)	0.12 (0.05–0.20)	133.16 (3.67–262.64)	0.95 (0.72–0.99)
CCE1 capsule Any polyp size (n = 8)	0.68 (0.61–0.74)	0.74 (0.68–0.79)	2.67 (2.07–3.26)	0.43 (0.34–0.52)	6.18 (3.59–8.78)	0.71 (0.71–0.88)
CCE2 capsule Any polyp size (n = 6)	0.90 (0.79–0.96)	0.81 (0.70–0.89)	4.81 (3.04–6.58)	0.11 (0.03–0.19)	44.47 (14.94–74.01)	0.82 (0.46–0.96)
CCE1 capsule size ⩾6 mm (n = 9)	0.69 (0.56–0.79)	0.83 (0.76–0.89)	3.78 (2.69–4.86)	0.40 (0.28–0.52)	9.45 (5.17–13.73)	0.73 (0.46–0.90)
CCE2 capsule size ⩾6 mm (n = 9)	0.88 (0.85–0.90)	0.92 (0.88–0.94)	9.03 (4.40–13.67)	0.13 (0.09–0.17)	68.32 (34.85–101.79)	0.92 (0.85–0.96)
CCE1 capsule size ⩾10 mm (n = 3)	0.90 (0.85–0.93)	0.88 (0.83–0.91)	7.30 (4.90–9.69)	0.11 (0.061–0.16)	65.75 (30.96–100.55)	NA- Insufficient number of studies
CCE2 capsule size ⩾10 mm (n = 11)	0.86 (0.75–0.93)	0.96 (0.94–0.97)	8.74 (5.16–12.31)	0.13 (0.056–0.21)	66.45 (18.34–114.56)	0.94 (0.71–0.99)

CCE, colon capsule endoscopy; DOR, diagnostic odds ratio; GLMM, generalised linear mixed model; NLR, negative likelihood ratio; PLR, positive likelihood ratio; SROC-AUC, summary receiving operator characteristic-area under the curve.

Table 3.

Relative sensitivity and specificity, and subgroup heterogeneity in sensitivity and specificity using GLMM and univariate meta-analysis.

Overall and subgroup	Parameter	GLMM		Univariate meta-analysis		Heterogeneity
		Estimate	95% CI	Estimate	95% CI	Random intercept variance (GLMM)	p	I2 (%)	τ2
Any Polyp size (n = 14)	Sensitivity	0.79	0.69–0.86	0.79	0.69–0.86	0.68	<0.001	77.9	0.67
	Specificity	0.77	0.71–0.82	0.76	0.56–0.90	0.19	0.003	58.3	0.18
Polyp size ⩾6 mm (n = 18)	Sensitivity	0.80	0.72–0.86	0.80	0.72–0.86	0.62	<0.001	84.3	0.62
	Specificity	0.87	0.82–0.91	0.87	0.82–0.91	0.60	<0.001	87.4	0.61
Polyp size ⩾10 mm (n = 14)	Sensitivity	0.88	0.79–0.94	0.88	0.78–0.94	0.99	<0.001	72.8	1.00
	Specificity	0.95	0.90–0.97	0.96	0.94–0.97	0.08	0.021	48.7	0.18
CCE1 capsule any polyp size (n = 8)	Sensitivity	0.68	0.61–0.74	0.68	0.61–0.74	0.06	0.004	66.5	0.08
	Specificity	0.74	0.68–0.79	0.74	0.68–0.80	0.02	0.056	49.1	0.09
CCE2 capsule any polyp size (n = 6)	Sensitivity	0.90	0.79–0.96	0.91	0.81–0.96	0.69	0.054*	23.9*	0.62
	Specificity	0.81	0.70–0.89	0.79	0.69–0.86	0.28	0.012	65.7	0.24
CCE1 capsule size ⩾6 mm (n = 9)	Sensitivity	0.69	0.56–0.79	0.69	0.56–0.79	0.41	<0.001	78.5	0.41
	Specificity	0.83	0.76–0.89	0.83	0.76–0.89	0.36	<0.001	79.5	0.36
CCE2 capsule size ⩾6 mm (n = 9)	Sensitivity	0.87	0.82–0.91	0.87	0.82–0.90	0.07*	0.23*	23.9*	0.07*
	Specificity	0.90	0.84–0.94	0.90	0.83–0.94	0.64	<0.001	91.0	0.65
CCE1 capsule size ⩾10 mm (n = 3)	Sensitivity	NA	NA	NA	NA	NA	NA	NA	NA
	Specificity	NA	NA	NA	NA	NA	NA	NA	NA
CCE2 capsule size ⩾10 mm (n = 11)	Sensitivity	0.86	0.75–0.93	0.86	0.74–0.93	0.88	<0.001	76.1	0.89
	Specificity	0.96	0.94–0.97	0.96	0.94–0.97	0.19	0.0082	56.6	0.19

Figure 2.

Forest plot of the sensitivity of polyp detection, stratified by polyp size categories using CCE.

CCE, colon capsule endoscopy.

Figure 3.

Forest plot of the specificity of polyp detection subgrouped into different polyp size categories using CCE.

CCE, colon capsule endoscopy.

The pooled per-patient sensitivity and specificity for polyps of any size were 0.79 (95% CI: 0.69–0.86) and 0.77 (95% CI: 0.71–0.82), respectively, with an AUC of 0.81 (95% CI: 0.47–0.96). For polyps ⩾6 mm, the pooled sensitivity, specificity and AUC were 0.80 (95% CI: 0.72–0.86), 0.87 (95% CI: 0.82–0.91) and 0.81 (95% CI: 0.48–0.95), respectively. Detection of clinically significant polyps (⩾10 mm) showed a sensitivity of 0.88 (95% CI: 0.79–0.95), specificity of 0.95 (95% CI: 0.92–0.97) and AUC of 0.95 (95% CI: 0.72–0.99). Due to the persistent heterogeneity, a meta-regression analysis was conducted using completion rate, bowel cleansing adequacy and the CCE capsule model as moderators. Only the CCE capsule model demonstrated borderline significance (p = 0.056).

Subsequently, a subgroup analysis comparing CCE2 to CCE1 showed a relative sensitivity of 0.76 (95% CI: 0.67–0.86, p < 0.001) and a relative specificity of 0.93 (95% CI: 0.81–1.07, p = 0.386). While CCE2 exhibited improved pooled sensitivity and specificity, only the improved sensitivity reached statistical significance. For CCE2, the pooled sensitivity and specificity for detecting polyps of any size were 0.90 (95% CI: 0.79–0.96) and 0.81 (95% CI: 0.70–0.89), respectively, with an AUC of 0.82 (95% CI: 0.46–0.96). In addition, CCE2 achieved an AUC of 0.92 (95% CI: 0.85–0.96) for detecting polyps ⩾6 mm and 0.94 (95% CI: 0.71–0.99) for polyps ⩾10 mm. Given that only three CCE1 studies reported data on polyps ⩾10 mm, the pooled estimates for this subgroup should be interpreted with caution, and AUC was not assessed. There was no statistically significant difference in diagnostic accuracy between polyps ⩾6 mm and polyps ⩾10 mm in CCE2. Even though substantial heterogeneity was observed in this meta-analysis, subgroup analysis revealed low heterogeneity in cases with high sensitivity for detecting polyps of any size, particularly at the ⩾6 mm threshold when using CCE2 (see Table 2).

In addition, the pooled sensitivity of detecting CRC using CCE was 0.96 (95% CI: 0.73–1.00), excluding cancer found in unseen colonic segments due to capsule battery exhaustion before excretion. A total of 81 patients were diagnosed with CRC. The forest plots illustrating the sensitivity of CCE for CRC detection when including and excluding cases from incomplete procedures are presented in Figure 4. The sROC curves of overall any-size polyp detection using CCE, any-size polyp detection using CCE2 and the ⩾6 mm polyp detection using CCE2 are shown in Figure 5.

Figure 4.

Forest plot of the sensitivity of colorectal cancer using CCE (a), including cancer found in the unseen colonic segment due to battery exhaustion before capsule excretion and (b) excluding cancer found in the unseen colonic segment due to incomplete procedure.

CCE, colon capsule endoscopy.

Figure 5.

Summary receiver operating characteristic (sROC) curves illustrating (a) detection of polyps of any size using CCE, (b) detection of any polyp size with CCE2 and (c) detection of polyps ⩾6 mm with CCE2, modelled using a generalised linear mixed model (GLMM) via the glmer function from the lme4 R package.

CCE, colon capsule endoscopy.

Discussion

Despite nine systematic reviews, including eight meta-analyses, confidence in CCE remains limited, hindering its widespread global adoption. This meta-review and umbrella meta-analysis is the first comprehensive synthesis of existing literature to reinforce trust and reliability in this technology. Our synthesis revealed that while CCE demonstrated a reasonable per-patient AUC of 0.81 when compared to colonoscopy as the reference standard, the second-generation model has significantly improved; notably, with CCE-2, the AUC for detecting polyps ⩾6 mm increased to 0.92 while maintaining a high AUC of 0.94 for polyps ⩾10 mm. This improvement was primarily driven by a statistically significant increase in sensitivity, with a relative sensitivity CCE1/CCE2 of 0.76 (95% CI: 0.67–0.86, p < 0.001). These advancements can be attributed to technological improvements in CCE, including enhanced camera and video resolution and extended battery life. This is also attributed to the evolution of improved bowel preparation, booster regimens and reading software over time.

Although heterogeneity was high in most pooled results, it is essential to note that CCE2 demonstrated a high sensitivity of 0.90 for any polyp size and 0.87 for polyps ⩾6 mm, both of which showed low heterogeneity (p > 0.05, I² < 25%). This suggests that the findings are reliable and consistent with some of the included systematic reviews. Furthermore, the expected trend of improved diagnostic accuracy with increasing polyp size is reaffirmed, as larger polyps are more readily identifiable.

The reduced specificity in smaller polyps primarily drives the lower accuracy observed for detecting polyps of any size. This can be attributed to two key factors. First, the false positive rate in CCE, particularly when mucosal folds, could be misinterpreted as small polyps due to partial visualisation of the fold within the frames. Second, even though colonoscopy is considered the gold standard, it is inherently imperfect, with a well-documented polyp miss rate of 26% (95% CI: 23–30) for polyps of any size and up to 26% for those ⩽5 mm. ⁴⁶ Given the absence of a perfect reference standard, accurately determining CCE’s factual specificity and diagnostic accuracy for detecting polyps of any size remains a challenge.

One of the included primary studies led by Kobae-Larsen et al. attempted to address this issue by performing tandem colonoscopies when significant lesions detected by CCE were initially missed on the first colonoscopy. This demonstrated that the per-patient sensitivity of CCE for detecting polyps ⩾9 mm was 97% higher than 89% for colonoscopy.^41,47 This suggests that the higher rate of missed lesions in colonoscopy may lead to overestimating CCE false positives, consequently underestimating its true diagnostic accuracy. Similarly, a study by Rex et al. ⁴⁸ identified 52 polyps detected by CCE but missed by colonoscopy, of which 22% were later verified through repeat colonoscopy. This further highlights the inherent limitations of colonoscopy as a reference standard.

For CRC detection, it is crucial to distinguish between true missed cancers and cases where the procedure was incomplete. In incomplete cases, the tumour was not overlooked but rather not visualised due to the incomplete procedure – conceptually similar to a colonoscopy with inadequate bowel preparation requiring a repeat examination. These cases should be classified as incomplete procedures rather than true missed cancers to ensure an accurate assessment of diagnostic performance.

When cancers that were missed due to incomplete procedures (e.g., capsule battery exhaustion before excretion) were included, the pooled sensitivity was 0.84 (95% CI: 0.68–0.93). On reviewing individual primary studies, Van Gossum et al. ⁴² did not specify whether the five missed CRCs were due to an incomplete procedure or true misses by CCE. Ota et al. reported true missed cases, while Kobaek-Larsen et al. documented four, all attributed to incomplete procedures.^31,41 Pecere et al. reported one missed CRC, which was detected but misclassified as an advanced polyp ⩽5 mm rather than malignancy. The only confirmed two missed CRC cases were reported by Sacher-Huvelin et al. ⁴⁹ When missed CRC cases from incomplete procedures are excluded, the pooled sensitivity improves to 0.96 (see Figure 4), with no heterogeneity, aligning with the previously reported 93% sensitivity by Vuik et al. ²¹ This further reinforces the potential role of CCE in bowel cancer screening, as outlined in the ESGE guidelines. ⁵⁰ Since most missed CRC cases were due to capsule battery exhaustion before excretion, optimising or personalising booster regimens is crucial to reducing the need for additional endoscopic procedures. Moreover, the true CRC miss rate in CCE remains an important area for future research, warranting further investigation. Unfortunately, CCE indications varied widely, including symptomatic and asymptomatic patients, with or without FIT or FOBT screening, and those with a family or personal history of CRC. This substantial heterogeneity precluded meta-regression and subgroup analyses.

Among the studies analysed, 94% utilised polyethylene glycol (PEG) as the primary bowel preparation (2–4 litres), while 81% used sodium phosphate-based laxatives as a booster. The wide variability in bowel cleansing adequacy (40%–90%) and capsule excretion before battery depletion (64%–100%) remains challenging, consistent with previously reported ranges. ⁵¹ Although Ohmiya et al. ⁵² proposed a castor oil and PEG combination to improve capsule excretion rates from 81% to 97%, and Deding et al. ⁵³ reported a prucalopride-driven increase from 57% to 75%, an optimal regimen has yet to be identified. These inconsistencies emphasise the urgent need to standardise and optimise bowel preparation and booster regimens. Interestingly, despite being widely considered key factors influencing accuracy, meta-regression analysis found no statistically significant association between completion rate, bowel preparation adequacy and polyp detection accuracy.

Despite the high sensitivity and diagnostic accuracy of second-generation colon capsule endoscopy (CCE-2) for polyp detection, cost-effectiveness remains a key barrier to its widespread adoption. ⁵⁴ This is largely influenced by the rate of follow-up endoscopies, as highlighted in a recent meta-analysis by Lei et al. ⁵⁵ A large-scale cost-effectiveness evaluation from the ScotCap study compared the total cost per CCE procedure (£747/€900) with conventional colonoscopy (£900/€1,085), based on National Services Scotland micro-costing data. CCE incurred an additional cost of £64.75 (€79.41) per patient in surveillance settings but yielded marginal savings of £6.71 (€8.23) per patient in symptomatic populations.^56,57 These savings were primarily due to a reduced need for colonoscopies and the reclassification of some urgent procedures to non-urgent flexible sigmoidoscopies.

One key limitation to consider is the challenge of polyp matching between CCE and colonoscopy. This difficulty arises from multiple factors. First, as Schelde-Olesen et al. ⁵⁸ demonstrated, unreliable localisation of anatomical landmarks can lead to inaccurate mapping of polyp locations in CCE. Second, CCE’s polyp sizing tool tends to overestimate lesion size, while polyp size assessment during colonoscopy is inherently variable between endoscopists, ⁵⁹ further complicating accurate matching. As a result, unmatched or incorrectly matched polyps may negatively impact CCE’s diagnostic accuracy, particularly when colonoscopy is assumed to be the infallible gold standard. This issue was reflected in Parodi et al., ⁶⁰ where patients were classified as false positives if CCE reported a polyp ⩾6 mm, but colonoscopy measured it as <6 mm. Furthermore, the absence of a truly accurate gold standard may underestimate the true sensitivity of CCE while simultaneously overestimating the specificity of colonoscopy, warranting further investigation into more robust validation methods. The absence of PROSPERO registration could be viewed as a limitation due to potential selection or reporting bias. However, this decision was rooted in the novelty of our methodological approach. At the time of study establishment, uncertainty existed regarding the eligibility of such a design for registration, and concerns arose about the potential for idea misappropriation upon public protocol disclosure. This was previously noted in registry analyses and surveys, which further informed this decision. ⁶¹ While we acknowledge this limitation, we followed a predefined internal protocol that remained consistent throughout the study.

Conclusion

This meta-review reaffirms the high sensitivity and accuracy of CCE2 in detecting polyps, particularly size ⩾6 and colorectal cancer. While challenges remain in optimising bowel preparation, booster regimens, capsule battery life and patient selection, CCE2 can potentially complement conventional endoscopy and alleviate endoscopy services’ workload. With AI integration on the horizon, polyp detection accuracy in CCE could potentially be further enhanced, paving the way for more efficient and scalable colorectal cancer screening solutions.